A Tunable Loss Function for Binary Classification
Tyler Sypherd, Mario Díaz, Lalitha Sankar, Peter Kairouz · 2019
We present α-loss, α ∈ [1, ∞], a tunable loss function for binary classification that bridges log-loss (α = 1) and 0-1 loss (α = ∞). We prove that α-loss has an equivalent margin-based form and is classification-calibrated, two desirable properties for a good surrogate loss function for the ideal yet intractable 0-1 loss. For logistic regression-based classification, we provide an upper bound on the difference between the empirical and expected risk for α-loss at the critical points of the empirical risk by exploiting its Lipschitzianity along with recent results on the landscape features of empirical risk functions. Finally, we show that α-loss with α = 2 performs better than log-loss on MNIST for logistic regression.